Testbed Support for GridPP member institutes [mailto:TB-
> [log in to unmask]] On Behalf Of Ewan MacMahon said:
> This is a small case in point of what's wrong with the information system -
> what matters to an external user is what service I'm offering, and how they
> can access it. That's what (most of) the storage-centric information we
> publish is about, and why the system works relatively well for storage. The
> OS version under a service is not part of the interface, and does not need
> exposing.
The information system is also used by EGI management, and for them it is useful to know things like OS and middleware versions. The fact that the information is present doesn't mean that an ordinary user has to do anything with it if they aren't interested.
> As similar problem afflicts the whole 'subclusters' idea of trying to advertise
> exactly what number of which model CPUs a site has in what configurations
> in which individual batches of worker nodes. No-one actually needs to know
> that - the service offered is a more generic "SL6/x86_64/cvmfs" service. As
> well as being a nightmare for a cluster that has any diversity of hardware at
> all, it copes extremely poorly with novel cluster designs, in particular anything
> with a cloud backend.
There has always been a difficult trade-off as to how much detail about computing systems is relevant, I don't think there's any right answer - we've always had long debates about this kind of thing and we always end up with some kind of compromise.
> It's sometimes worth bearing in mind the philosophical grounding of why we
> call this a grid at all - it's supposed to work like the power grid, and when I
> plug in an appliance I don't know whether the electricity was generated by
> coal, gas, nuclear, wind, or the solar panels on some guys house.
Indeed, but that mainly shows that the power grid analogy was always fairly meaningless.
Stephen
|